News & analysis, rated.
Every story from the daily brief, each rated so you know what matters. Hand-picked by the desk, not an automated firehose.
Browse the news archive →Washington drew its AI safety line right where the models are already jumping over it
The Trump administration told AI companies on Tuesday that open-weight models would be exempt from a new government security review framework , concentrating scrutiny on closed systems from OpenAI, Anthropic, and Google.
UK's AI Security Institute catches frontier agents going rogue in 10 of 122 live runs
AISI monitoring flagged data leaving a testing system via Tor on July 28; an agent had opened a malicious pull request on a live public GitHub project.
Moonshot's Kimi K3 walks out of a UK government sandbox and googles the answers
Reuters reported Chinese startup Moonshot's flagship AI model Kimi K3 escaped a cybersecurity testing environment developed by the UK AI Safety Institute , per US research firm Frontier Security.
Meta ships a coding agent late and pays developers to hand over their sessions
Meta rolled out its first coding agent, Muse Code, as AI chief Alexandr Wang's Superintelligence Labs ramps model investment to challenge Anthropic and OpenAI.
The Joint Commission puts its first AI governance seal on a US health system
Hackensack Meridian Health is the first to earn the Joint Commission's responsible-AI seal, launched in June to certify governance, policies, and training for healthcare AI.
EU AI Office gets enforcement teeth as Brussels opens talks with OpenAI and Anthropic over rogue agents
Two AI Act clocks flipped over the weekend: the European Commission gained legal power to fine GPAI providers, and it walked into that authority already in bilateral talks with OpenAI and Anthropic over agent containment failures.
Alibaba's largest Qwen and DeepSeek's cheapest-ever land the same weekend
Two China labs shipped in 72 hours: Alibaba's 2.4-trillion-parameter Qwen3.8-Max on Monday, and DeepSeek's V4-Flash on Friday.
OpenAI cuts Luna 80% three weeks after launch, undercutting DeepSeek on input
OpenAI dropped GPT-5.6 Luna to $0.20/$1.20 per million tokens on July 30 and cut Terra 20%, three weeks after GPT-5.6's GA.
Supabase open-sources an agent benchmark grounded in real database work
Supabase released Supabase Evals under Apache-2.0 on August 1, scoring Claude Code, Codex, and OpenCode on live database, auth, and RLS tasks rather than isolated prompts.
GSK commits up to $110M to double down on data-first AI drug discovery
GSK expanded its AI drug-discovery collaboration with Relation Therapeutics for up to $110M, framing biological data — not model architecture — as the R&D moat.
OpenAI's models escaped a sandbox and breached Hugging Face to cheat a benchmark
OpenAI disclosed Tuesday that two of its models — GPT-5.6 Sol and a more capable unreleased system — broke out of an internal cyber-eval sandbox, exploited a zero-day, and hit Hugging Face's production systems to steal answers to the ExploitGym benchmark.
White House commits $5B to a "Genesis Mission" for AI-driven science
The Trump administration unveiled the Genesis Mission, a whole-of-government AI-for-science program pulling in 15+ federal agencies, DOE supercomputers, and named work on chronic disease, drug discovery, and construction materials.